Skip to content

Claude Haiku 5.5 (effort default (no --effort flag) · thinking on) on Catch log checked against a made-up fishing regulations table

AClaude Haiku 5.5default · thinking on · Claude Code 2.1.299 (headless -p) · cloud · 2026-10-20 · by @bobsiopener

Trust 80%Fully autonomoussuccess · first try
Estimated cost
$0.0069
Wall time
27s
Human step-ins
0
Outcome
Pass

Compare with other models on this task →

Summary

Exact prompt

# Catch log checked against a made-up regulations table `input.json` names `regulations.json` and `catch_log.csv`. All rules, species codes and data are **fictional** (not real fishing regulations). Write `solve.py` (Python 3 standard library only); `python solve.py input.json answer.json`. Fresh files are used too. ## Data - `regulations.json`: `species` maps a code to `aliases`, `season` ["MM-DD", "MM-DD"] inclusive (if start > end the season wraps over New Year), `min_in`, `max_in` (null = no maximum), `bag` (legal kept fish per angler per day for that species); `aggregate_daily` = legal kept fish per angler per day, all species. - `catch_log.csv`: `date` (YYYY-MM-DD), `angler`, `species`, `length`, `unit` (`in` or `cm`, any case), `action` (`kept` or `released`, any case). Process rows in file order; row numbers start at 1 for the first data row. ## Rules (first match wins) 1. Species: trim, case-insensitive match to a code or alias; no match -> `unknown_species` (species null). 2. `released` -> status `released` (never a violation, never counted). 3. Kept fish: length in inches (cm / 2.54 rounded to 2 decimals). Then: not in season -> `out_of_season`; below min -> `undersize`; above max -> `oversize`; angler already has `bag` legal fish of that species that date -> `over_bag`; angler already has `aggregate_daily` legal fish that date -> `over_aggregate`; else `legal` (and it counts toward the limits). Boundaries are legal (equal to min or max, first/last season day). 4. Every status except `legal` and `released` is a violation for that angler. ## answer.json `{"rows": [{"row", "species", "status"}, ...], "violations_by_angler": {angler: n} (only anglers with >0, sorted by name), "legal_kept_by_species": {code: n} (only >0, sorted)}` ## Scoring Hidden checks (partial credit; all must pass): `species_and_release` (species codes and which rows are released), `row_statuses` (every row exact), `summaries`, `fresh_run_1`, `fresh_run_2`. 60-second limit per run.

Aliases with odd spacing and case, cm conversion near limits, a season that wraps New Year with boundary dates, slot maximum, and bag/aggregate limits that only count legal fish in log order. Original benchmark task written for usecaserank.com (not based on a specific public post). Headless, offline run on a reproducible fixture. Each run's notes say how its cost was measured (estimates are marked).

Benchmark code for comparing AI models, not navigation or safety software.

Success criteria: 1. species_and_release 2. row_statuses 3. summaries 4. fresh_run_1 5. fresh_run_2

Setup notes: Use Case Scout batch 10 case k04-catch-log-regulations (offline fixture). Prepared with batch10/prepare.sh; hidden key/tests kept outside the agent's work dir. Prompt = TASK.md verbatim. Scored with the scout's scorer: `python score.py WORKDIR KEYDIR`. Tools: Read/Write/Edit/Glob/Grep/Bash, web disabled. Single run, no retries. Claude Code's own reported cost for claude-haiku-5.5: $0.006899. Score: 5/5. Cost is an ESTIMATE: measured tokens x pricing.json list price (Claude Pro subscription, nothing billed); input_tokens include cache reads and writes. Single run.

Estimated cost List-price estimate; not actually billed.

Metrics

Estimated cost
$0.0069
Wall time
27s
Human step-ins
0
Turns
5
Model usage
ModelInput tokOutput tokCost
AClaude Haiku 5.5anthropic/claude-haiku-5.5effort default (no --effort flag) · thinking on92,7265,964$0.0069

Reproduce this run

  1. Use the use case prompt on the use case page.
  2. Same setup: Claude Haiku 5.5 (anthropic/claude-haiku-5.5, effort default (no --effort flag) · thinking on) via Claude Code 2.1.299 (headless -p), cloud.
  3. It counts as a success only if it meets every success criterion: 1. species_and_release 2. row_statuses 3. summaries 4. fresh_run_1 5. fresh_run_2
  4. Log model, tokens, wall time, human step-ins and the evidence log, then submit it as a reproduction. A different model? Submit a head-to-head rerun.

How runs are scored: Methodology.

Was this useful?

⇄ Rerun on another model

Automated checks

Trust 80% comes from these checks (how Trust % works).

Evidence log
[
 {
  "duration_api_ms": 26144,
  "stop_reason": "end_turn",
  "session_id": "[REDACTED]",
  "total_cost_usd": 0.00689887,
  "usage": {
   "input_tokens": 8,
   "cache_creation_input_tokens": 15731,
   "cache_read_input_tokens": 76987,
   "output_tokens": 5964,
   "output_tokens_details": {
    "thinking_tokens": 3299
   },
   "server_tool_use": {
    "web_search_requests": 0,
    "web_fetch_requests": 0
   },
   "service_tier": "standard",
   "cache_creation": {
    "ephemeral_1h_input_tokens": 15731,
    "ephemeral_5m_input_tokens": 0
   },
   "inference_geo": "global",
   "iterations": [
    {
     "input_tokens": 2,
     "output_tokens": 1962,
     "cache_read_input_tokens": 24136,
     "cache_creation_input_tokens": 4134,
     "cache_creation": {
      "ephemeral_5m_input_tokens": 0,
      "ephemeral_1h_input_tokens": 4134
     },
     "type": "message"
    }
   ],
   "speed": "standard",
   "fallback_credit": null
  },
  "modelUsage": {
   "claude-haiku-5-5": {
    "inputTokens": 8,
    "outputTokens": 5964,
    "cacheReadInputTokens": 76987,
    "cacheCreationInputTokens": 15731,
    "webSearchRequests": 0,
    "costUSD": 0.00689887,
    "contextWindow": 1000000,
    "maxOutputTokens": 128000,
    "thinkingTokens": 3299,
    "canonicalModel": "claude-haiku-5-5",
    "provider": "firstParty",
    "costBasis": "list"
   }
  },
  "permission_denials": [],
  "terminal_reason": "completed",
  "fast_mode_state": "off",
  "fast_mode_disabled_reason": "sdk_opt_in_required",
  "subagent_stats": {
   "spawned": 0,
   "requested": {
    "background": 0,
    "foreground": 0,
    "unset": 0
   },
   "started_in_background": 0,
   "max_depth": 0,
   "spawned_by_subagents": 0,
   "completed": 0,
   "failed": 0,
   "killed": {
    "parent": 0,
    "user": 0,
    "system": 0
   },
   "refused": {
    "depth_limit": 0,
    "concurrency_limit": 0,
    "budget": 0
   },
   "by_type": {}
  },
  "safety_stops": 0,
  "is_error": false,
  "num_turns": 5,
  "subtype": "success",
  "api_error_status": null,
  "result": "I wrote `solve.py` in the work directory. It runs on the sample input in about a second and writes `answer.json` in the format the spec describes. I have no hidden-check results, so correctness beyond hand-spot-checks is unverified.\n\nRun it with `python3 -I solve.py input.json answer.json`. Sample output: 55 rows, violations for ana (15) and ben (12), and legal kept counts of BSB 5, FLK 3, STB 5, TOG 6.\n\nChoices I made where the spec was open:\n- **Bag and aggregate limits** count legal kept fish per angler per calendar date, only the ones already marked `legal`. Out-of-season, undersize and oversize fish do not count.\n- **Unknown species** rows are a violation under rule 4, so they count in `violations_by_angler`. Released rows never count.\n- **Centimetres** are converted with `round(cm / 2.54, 2)` and inches are used as given. Limit comparisons are inclusive.\n- **Season wrap** works as `date >= start or date <= end` when start > end (e.g. tautog 10-15 to 01-31).\n- **Matching** trims the species text and compares it case-insensitively to codes and aliases. Internal spacing is not normalised. Angler names are trimmed but otherwise kept as written.\n- **Relative paths** in `input.json` are resolved against that file's directory, falling back to the current directory.",
  "ttft_ms": 1117,
  "type": "result",
  "duration_ms": 26635,
  "uuid": "[REDACTED]",
  "ttft_stream_ms": 987,
  "time_to_request_ms": 353,
  "first_content_frame_ms": 987,
  "queued_turn_count": 0,
  "result_index": 0
 }
]

Get new runs and ranking changes weekly